This paper provided a new industrial nonpoint source reallocation method, a NPS method, based on blue-roof industrial buildings based on satellite imagery. The results indicate using the NPS method will improve the model performance compared with the conventional method of using population density. In general, the paper is clearly organized and easy to follow, and provided a new spatial allocation emission data for the Greater Bay Area (GBA) area. However, the authors mainly compared this method with the regional emission inventories using top-down spatial allocation methods. Currently, the high-resolution emission inventories of most key regions in China adopt point-source processing based on "bottom-up" method. I encouraged the authors to expand the scope of the paper and to discuss more with the point-based high-resolution emission inventory development. Some more detailed information should also be provided and discussed, before the paper can be accepted for final publication. Details follow.

1. As mentioned above, my main concern is that the discussion could be expanded a little bit, and would be focused more on the comparison with point-source based spatial allocation method. That might be more helpful for the whole research community. It invites more review on published work for the high-resolution regional emission inventory, and more comparison and discussion with these inventories.
2. I can understand that using the NPS method can effectively improve the spatial allocation of emissions. However, it seems that the method should not be able to distinguish the differences in emissions between the factories. After all, using images cannot accurately determine the scale of different companies. If it cannot be distinguished, how much uncertainty was caused by this?
3. The advantage of this method is to improve the spatial distribution of emissions. However, this study only selected limited city (like Zhongshan in Figure 7) when verifying the model performance. I suggest more monitoring sites should be included to reflect the advantages of improved spatial distribution.
4. In “3.4 conclusion remarks”, the following sentences seem to be unnecessary repetitions of the body content, and it is recommended to delete them.
5. There are some mistakes in the MS as well. (1) Line 49. “population density wasa applied as” should be “population density was applied as”. (2) Line 134. “In thi study” should be “In this study”.

The authors developed a new method to allocate industrial emissions onto grid cells based on the areas of blue-roof buildings, which are retrieved from the satellite imagery. This new emission distribution method has been applied to the MIX inventory and evaluated through atmospheric chemistry modeling and comparison against surface observations. Overall, I feel that this study provides new insights into high-resolution emissions mapping, which deserves publication. It has been recognized that the population-based allocation method tends to overestimate anthropogenic emissions over urban areas in China, which could be improved using the method developed in this work to identify the location of blue-roof industrial buildings. My only concern is that the manuscript lacks a detailed description of the blue roof identification algorithm, which is difficult to understand in its current form. And the evaluation results using surface observations need further analysis to illustrate the improvement of the spatial distribution patterns of emission inventories. My comments are as follows.

• 3.1. The description of the emission allocation method should be moved to the method part in Section 2. After reading this section, I am not very clear how the blue roof identification algorithm works. What do you mean by “incorporated the effect of sun position on colour change under different latitudinal position in the satellite images.” (lines 175-176 on page 5)? Please explain “In this study, 4 ranges of HSV were identified for the blue roof identification algorithm. Its HSV ranges were 193-230° for H, 17% to 90% for S and 40% to 100% for V.” (lines 179-180 on page 5)? The method description needs to be further improved to make it easier for an audience to understand.
• 3.3. The evaluation of the CMAQ simulation only presents the summary of performance statistics that covers all of the surface observation stations. I am curious whether the model performance (e.g., RMSE, MB) is different among urban, rural, and remote background observation stations, which will help understand the improvement of emission distribution patterns over different regions.
• Uncertainty assessment. Table 2 presents the False Detection Rate and False Alarm Rate in the blue roof identification algorithm. Is it possible to incorporate this information to quantify the uncertainties in the emission allocation processes?
• Line 105, page 4. Please explain how the MIX inventory for 2010 was scaled to the target simulation year of 2015.
• Line 108, page 4. “Hong Kong emissions were not presented in the MIX inventory” If I remember it correctly, the MIX inventory includes Hong Kong emissions.
• Lines 289 and 290, page 8. “However, we are also aware that the assumption of the blue-roof method where larger blue-roof has more emissions may not always be sufficient under different resolutions.” This looks very interesting, but the manuscript has not analyzed the emission distribution patterns at different spatial resolutions, right? I would like to know how the authors reach such a conclusion.